Visual Pose Estimation And Model Predictive Control For Rendezvous And Docking With A Non-cooperative Target

Since the increasing non-cooperative targets occupy too much space resource,and the chaser spacecrafts are usually required to have autonomous navigation and control capabilities,this dissertation studies the autonomous rendezvous and docking strategy with a tumbling non-cooperative target,which mainly includes monocular relative pose estimation and model predictive control strategy.The specific research contents are as follows:The residual convolutional neural network is adopted to estimate the relative pose between the mission spacecraft and a non-cooperative target.Based on the dataset which is produced by Stanford SLAB,with the Tango satellite as the estimated target,the nonlinear transformation between the monocular gray image of the target and its 6-DOF pose information is fitted.Consider the coupling between the position and attitude of the target,a dual-branch residual neural network is proposed to decouple.In addition,since the relative position is described by the coordinates between the chaser’s and the target’s centroid,the projection of the three-dimensional centroid in the two-dimensional plan is affected by the attitude of the target and its shape,which results the deep learning of the relative position is hader than the relative attitude.Therefore,a 50-layer convolutional ResNet-50 which is more suitable for complex models for the relative position learning is adopted,a 34-layer convolutional ResNet-34 for the relative attitude learning is adopted.Finally,experimental data verify the effectiveness of the proposed strategy.The output of the estimator is the input of the controller,and the basis of control is a precise and practical position-attitude dynamics framework.Consider the advantages of LOS dynamics and Euler dynamics,that the navigation information can be directly used without any coordinate transformation.Since existing papers have not considered the two dynamics in one framework,a novel frame called LOS-Euler framework is proposed by defining the way of Euler rotation from the reference coordinate system to the LOS frame and the body frame with a same rotation sequence,which can effectively describe the coupling between the position and attitude of the tumbling target’s docking port.In addition,consider the constraints on the input saturation,collision avoidance constraints,entry cone,field of view,the existing constraints construction method does not apply to the proposed LOS-Euler framework,hence the constraints are reconstructed.The novel construction strategy can reduce the effort of calculation and avoiding the linear approximation.In addition,since the Euler angles is adopted to describe the attitude,there will be two mathematical expressions corresponding to the same angular position,which is called singularity.Consider a tracking problem,the reference signal will jump once it reaches a singular point.Therefore,a Singularity-free strategy is proposed to avoid this situation.The control strategy of RVD with a non-cooperative target is also studied in this dissertation,model predictive control(MPC)is adopted for the advantage of dealling with constraints.Since the dynamics in LOS-Euler framework are nonlinear,the piecewise affine model(PWA)is adopted.The standard PWA-based predictive control is firstly designed and simulated to verify the effectiveness of the proposed strategies.Subsequently,the PWA-based PC is analyzed,and a PWA-based optimal predictive control(PWA-based OPC)strategy is proposed.Then,based on PWA-based OPC,a novel PWA-based improved optimal predictive control(PWA-based IOPC)is proposed and the linear matrix inequality(LMI)and S process theory are used to prove the stability of the proposed algorithm.Consider the error of the relative pose estimator,this dissertation integrates the controller and the filter.Finally,a tube-based robust predictive control strategy is introduced and the stability is improved.Numerical simulation shows that the PWA-based IOPC controller proposed has improved the control performance,and the strategies can achieve ideal results.